Lungs CT Data

Dataset Description

Context

Competitions like LUNA (http://luna16.grand-challenge.org) and the Kaggle Data Science Bowl 2017 (https://www.kaggle.com/c/data-science-bowl-2017) involve processing and trying to find lesions in CT images of the lungs. In order to find disease in these images well, it is important to first find the lungs well. This dataset is a collection of 2D and 3D images with manually segmented lungs.

Preprocessing

We can convert the TIFF images to JPEG files and copy them into a new directory.

Train and Test sets

Modeling: Modified U-Net

We use a modified U-Net. A U-Net consists of an encoder (downsampler) and decoder (upsampler). We load and modify an exiting MobileNetV2 model from tf.keras.applications

The upsample block is already implemented in TensorFlow Examples in the Pix2pix tutorial.

Compiling the model

Fitting the model

Removing converted jpg images.


References

  1. Kaggle Dataset: Satellite Images of Hurricane Damage
  2. Tensorflow API Documentation
  3. Image segmentation